Micro-functional protein complexes mining in biological intelligent computing: a weighted network approach

Identifying protein complexes is of great significance for drug target discovery and understanding disease mechanisms. Recognizing the variations of different protein complexes in individual organisms aids in the development of personalized treatment strategies. In order to identify potential protein complexes with distinct modularity and density, as well as overlapping protein complexes, we propose a method called Micro-Clusters Overlap Reconstruction (MCOR) to mine micro-functional protein complexes by reducing the complexity of protein complexes. The method, by incorporating both network topology and protein biological information, can significantly reduce the impact of false-positive interactions in the protein network. Firstly, we create a weighted network based on functional annotation terms and shared neighbors. Secondly, we define a protein selection mechanism to form initial clusters. Thirdly, we define a complex evaluation function to identify complexes in the network with varying modularity and density. Fourthly, we design a seed expansion algorithm that utilizes the complex evaluation function to expand clusters and form complexes. On real data sets from multiple species, we compared MCOR with the currently most advanced seven algorithms. Our findings suggest that MCOR surpasses these algorithms in terms of F \(_{1}\) -measure and accuracy criteria across networks generated from different species’ data. Moreover, the identified complexes have a low average p-value, thereby confirming the authenticity of the complexes identified by MCOR.